# Minimax Code Trajectory

> Inspect or visualize privacy-aware timelines for local MiniMax Code sessions. Use when the user asks to show, draw, open, review, or diagnose a task trajectory, summarize tool or token activity, inspect compaction or failure events, or compare recent session metadata without opening raw ledger files.

- Skill: `minimax-ai/minimax-code-trajectory` (Agent Skill)
- Install (CLI): `npx skillmds@latest add minimax-ai/minimax-code-trajectory`
- Raw SKILL.md: https://api.skillmd.com/api/skills/minimax-ai/minimax-code-trajectory/raw
- Safety review: pending
- Works with: Claude Code, Claude.ai, OpenAI Codex
- Category: Research & Search
- License: Apache-2.0
- Author: MiniMax AI (https://skillmd.com/u/minimax-ai)
- Updated: 2026-09-17
- Page: https://skillmd.com/skills/minimax-ai/minimax-code-trajectory

---


# MiniMax Code Trajectory

Use the bundled read-only MCP tools to inspect MiniMax Code session ledgers. Do not search the
filesystem manually when the tools can answer the request.

## Workflow

1. If the user asks to show, draw, open, or visualize a trajectory, call
   `show_minimax_trajectory` with `detailLevel: "summary"` and a bounded `maxRecords` value. Omit
   `sessionId` for the latest readable session unless the user selected another session.
2. Read the exact `visualization.fileUrl` from the tool result. If the MCode built-in Browser is
   available, call Browser with action `navigate` and pass that exact file URL. Do not use shell
   commands, an external browser, or a guessed path. After Browser succeeds, tell the user the
   interactive trajectory is open; do not dump the trajectory JSON into the conversation.
3. If Browser is unavailable, return the `visualization.fileUrl` as the fallback and explain that
   the generated self-contained HTML can be opened in MCode Desktop. Do not claim it was opened.
4. If the user asks for analysis without a visual page, call `get_minimax_trajectory` instead and
   report the event sequence, session state, failures, message/tool counts, compactions, token
   usage, and parser warnings that answer the question.
5. Distinguish observed ledger facts from inference. A missing event does not prove an action never
   happened outside the retained ledger.

## Privacy boundary

- Use `summary` by default. It intentionally omits conversation text, thinking, tool arguments and
  results, titles, absolute paths, file identifiers, hashes, credentials, and raw records.
- Use `detailLevel: "full"` only after the user explicitly asks to inspect message content in this
  conversation. State that bounded redacted previews will enter the active conversation before the
  call.
- Never claim `full` is a raw export. It still omits thinking, tool arguments/results, attachments,
  raw metadata, and absolute paths.
- Do not ask for credentials or broaden the data directory. If the desired profile is not found,
  tell the user to launch MiniMax Code with the matching `MINIMAX_DATA_DIR` or `MAVIS_DATA_DIR`.

## Failure handling

- `sessions_root_missing`: the selected profile has no local-runtime v2 sessions yet, or the wrong
  data directory is active.
- `session_not_found`: list recent sessions and ask the user to choose an available ID.
- `ledger_missing`: the manifest still exists but its trajectory ledger has been removed; choose a
  session returned by `list_minimax_sessions` instead.
- `malformed_jsonl_line`: report that a complete ledger record was corrupt; do not invent it.
- `incomplete_tail_ignored`: the session may still be writing; explain that the final partial record
  was ignored safely.
- `oversized_jsonl_line`: a record exceeded the 2 MiB safety cap and was omitted. Do not treat the
  returned trajectory as complete.
- `symlink_artifact_rejected`: stop. Do not follow or read the symlink target.

## Output style

Lead with the result, then include the session ID, time range, high-signal event counts, failure or
compaction facts, tool/token observations, and warnings. Keep raw JSON out of the answer unless the
user explicitly requests it.

